知识卡片
匿名化与假名化
降低数据访问与审计面的常见做法是匿名化。较简单的是假名化(pseudonymization):用可逆方式把私有标识替换为其他标识,需额外数据/系统才能还原,可防工程师顺手查看,但应视同未匿名数据一样高风险——历史上有大量借“匿名”数据反推真实个人的案例。真正匿名需永久切断人与训练数据的直接关联(如按聚合使单条数据不对应少于某数量真人)。Sweeney 证明仅凭性别、年龄、邮编即可唯一识别 87% 美国人。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
原书出处 · 1 处
- Data Management Principles · 查看原书依据
Sweeney further demonstrated that 87% of all people in the US could be uniquely identified via only three pieces of information: gender, age, and zip code; see “Anonymized Data Really Isn’t—and Here’s Why Not” by Nate Anderson for more details. Similar work has shown that a browser user agent string combined with zip code or several other kinds of information uniquely identifies humans.
参考来源
- 位置:《Reliable Machine Learning》第02章《Data Management Principles》(源文件:extracted/Reliable Machine Learning/_epub-src/OEBPS/ch02.xhtml)
- 结论依据:原文区分假名化与匿名化,并以 Sweeney 研究说明匿名化之难与语境依赖。
- 原始内容:Sweeney further demonstrated that 87% of all people in the US could be uniquely identified via only three pieces of information: gender, age, and zip code; see “Anonymized Data Really Isn’t—and Here’s Why Not” by Nate Anderson for more details. Similar work has shown that a browser user agent string combined with zip code or several other kinds of information uniquely identifies humans.